You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Something will come. I've almost finished an AI add-in for Docspell that can use the OpenAI API, or potentially other LLMs, to help suggest missing tags for documents.
The idea is to work with Docspell's existing tagging rather than replace it. The add-in first looks at the tags you already have in Docspell and uses those where they are appropriate. If it thinks a useful tag is missing, it can ask the LLM to suggest one.
New tags that you accept are then saved normally in Docspell. This means Docspell can learn from those tags as well, so over time it should become better at tagging documents itself and need less help from the LLM. That should also keep API usage and cost down.
I've been putting quite a bit of effort into keeping the number of tokens sent to the LLM as low as practical, with configurable input and output token limits. API calls are generally very cheap, but there's no point sending more document content than is needed just to work out some tags.
It can also show you its proposed tags before applying them. This should help avoid ending up with hundreds of slightly different AI-generated tags and let you keep your tag structure organised.
It works through Docspell's normal add-in system, so you can choose to run it only on documents you want AI help with, or have it run automatically as documents are added.
Setup is fairly simple for an existing Docker Docspell installation. Add-ins need to be enabled in the Docspell YAML, and the API key can be added to the Joex environment. Things such as the model and token limits can then be set in the add-in arguments.
I'm still finishing and testing it, but I'd be interested to know how much interest there is in something like this. If other Docspell users would find it useful, I'll put some effort into making the setup and documentation straightforward for others to use.
It runs alongside Docspell and uses the Docspell add-in system together with a Companion web app. It reuses existing Docspell metadata where possible, can suggest missing metadata using AI, provides an approval workflow for new suggestions, and includes configurable rules, bulk scanning, scan history and diagnostics.
v1.0.0 currently uses OpenAI. Installation, user and developer documentation are all included in the repository/release.
I’d be interested in any feedback from other Docspell users who try it.
Activity
shanems-cell commentedon Aug 29, 2026
Something will come. I've almost finished an AI add-in for Docspell that can use the OpenAI API, or potentially other LLMs, to help suggest missing tags for documents.
The idea is to work with Docspell's existing tagging rather than replace it. The add-in first looks at the tags you already have in Docspell and uses those where they are appropriate. If it thinks a useful tag is missing, it can ask the LLM to suggest one.
New tags that you accept are then saved normally in Docspell. This means Docspell can learn from those tags as well, so over time it should become better at tagging documents itself and need less help from the LLM. That should also keep API usage and cost down.
I've been putting quite a bit of effort into keeping the number of tokens sent to the LLM as low as practical, with configurable input and output token limits. API calls are generally very cheap, but there's no point sending more document content than is needed just to work out some tags.
It can also show you its proposed tags before applying them. This should help avoid ending up with hundreds of slightly different AI-generated tags and let you keep your tag structure organised.
It works through Docspell's normal add-in system, so you can choose to run it only on documents you want AI help with, or have it run automatically as documents are added.
Setup is fairly simple for an existing Docker Docspell installation. Add-ins need to be enabled in the Docspell YAML, and the API key can be added to the Joex environment. Things such as the model and token limits can then be set in the add-in arguments.
I'm still finishing and testing it, but I'd be interested to know how much interest there is in something like this. If other Docspell users would find it useful, I'll put some effort into making the setup and documentation straightforward for others to use.
shanems-cell commentedon Sep 12, 2026
I’ve now published the first stable release of the project I mentioned above: Docspell AI Companion v1.0.0.
https://github.com/shanems-cell/docspell-ai-companion
Release:
https://github.com/shanems-cell/docspell-ai-companion/releases/tag/v1.0.0
It runs alongside Docspell and uses the Docspell add-in system together with a Companion web app. It reuses existing Docspell metadata where possible, can suggest missing metadata using AI, provides an approval workflow for new suggestions, and includes configurable rules, bulk scanning, scan history and diagnostics.
v1.0.0 currently uses OpenAI. Installation, user and developer documentation are all included in the repository/release.
I’d be interested in any feedback from other Docspell users who try it.
shanems-cell commentedon Sep 12, 2026
For clarity, v1.0.0 currently only supports OpenAI/ChatGPT models. Support for other LLM providers is something I’d like to add in a future version.